Software Development Effort Estimation Using UML Activity Models with Regression Analysis

Prediction of development effort of software is an important prerequisite for its actual development. However, the complexities involved in the creation process make it a stiff challenge to make viable prediction that is adequately precise. This study reveals a smart estimation approach for the present-day applications in various domains. The applied approach first extracts the details represented in the Unified Modeling Language (UML) Activity models. These details are then fed to a number of regression analysis procedures written for this study that includes: ridge (LRR), lasso (LLR), support vector (SVR), extreme gradient boosting (XGBR), decision tree (DTR), K-nearest neighbors (KNNR) and Bayesian ridge regression (BRR). The findings from the experimentation suggested that the BRR delivered a superior accuracy in train-test split as well as fivefold cross-validation.